The State of English Learner Education in the U.S.

Project Purpose and Goals

The following tables and visualizations are intended to give a broad overview of the programs, resources, and services high schools provided to English learner-classified students and their families as reported by the questionnaire responses. The primary objective for this project was to practice beginning functional programming skills, and thus the majority of the effort was placed in creating the tables and visualizations rather than on ensuring sound or thorough analysis.

The project utilized both descriptive approaches to the data to summarize how services and programs differed across regions and locale types, and quantitative approaches through the use of simple regression models. The questionnaire was deployed to central, western, northeastern, and southeastern United States and thus the original coding schemes were used to guide analysis. In some cases, locale types were modified from the original categories (city, suburban, town, rural) to a dichotomous urban/rural category for ease of interpretation.

I imagined a research team as the audience for this project, meaning little interpretation is offered as it’s meant to be a springboard for discussion and further analysis.

Other things to note

In order to conduct analyses, I created a set of scores that would provide me with the necessary continuous variables to compare regional and locale-based differences for this dataset.

These include scores for:
* instructional services and programs;
* language supports offered;
* non-instructional services offered;
* high school and parent services offered (including translation and interpretation services);
* and an overall score for resources offered.

In addition, I created a bilingual content score which created a composite score based on the type of services and programs that were offered to EL-identified students without penalizing monolingual approaches. In other words, rather than creating a score that indicated simply whether a school offered a given service/program or not, the bilingual score differs by prioritizing bilingual approaches for a full score of “1” and a half score for English-based services and programs.

All scores were pulled and created from the corresponding subgroups in the questionnaire:
* demographics
* instructional language services
* additional language supports
* non-instructional language services in the most common language for ELs
* services available for ELs and their parents/guardians

For more information, please seek out the source code (data-clean.Rmd and/or analysis.Rmd) which includes notes on variable manipulation.

Table 1

The following table is intended to describe how different services designed for EL-identified students are distributed across regions of the US. The score shown in each column is a combined average score for each region within each given composite score.

I anticipated that the average number of services would vary by region, which seemed to be the case, though to a much lesser degree than expected.

Unfortunately, across all regions the average services and programs appeared to be much lower than the maximum possible score. This may be due to error in creating composite scores, as it’s very likely schools will choose certain programs over others rather than utilizing multiple types of programs and services. In a future iteration of this analysis, I would more carefully select criteria for composite scores as informed by practice.

Table 2

Similar to the table above, the following table is meant to demonstrate how different services are distributed with locale as the focal variable.

Table 3

The final table considers the relationship between both region and locale, with the idea in mind that more differences would appear at a more granular level and that the differences may tell us something about how urban and rural schools differ in their approaches to providing services and programs for English learner-identified students.

Indeed, we are able to see some differences in scores with much higher scores for overall resources provided in western, central, and northeastern cities. (As a side note, it’s surprising that this dataset did not include the southwestern United States as it houses the highest concentrations of English learners.)

In a future iteration of this project, I would love to look at the specific services and programs offered rather than the combined score, as this seems to obscure some meaningful data that I did not have enough time to look at.

Visualizations Across Regions and Locale

I was intrigued by the differences this plot brought to the surface, and dug further into this area by comparing the top differences across all groups. I considered pulling top 5 differences for each region and locale, but I was curious if one region would be more heavily represented than the others.

This did, in fact, happen to be the case.

Differences between urban and rural areas were more pronounced in the Northeast, making up 4 of the total top 5 differences. The final top score was in a central region. It may be interesting to dig further into this finding to explore why language-focused supports, instructional programs and models, high school student and parent services, and overall bilingual offerings differed so much between urban and rural areas in the Northeast.

It’s worth noting that the composite score for high school and student and parent services showed up in 2 of the 5 total top differences. This score was very similar to the non-instructional language services score that included information about academic and CTE programs, written translations, and interpreter services for students and their families. This score differed in that it focused on other non-English languages represented in the school that were not the predominant non-English language. This is very telling about the services that may be available for speakers of other languages as they made up the 2 largest differences overall. This may in part be due to representation, as urban areas often show higher diversity in non-English language composition, but I am hesitant to make that speculation as it obscures the potential diversity of non-English languages spoken in rural areas–something which happens all too frequently as rural areas are (often incorrectly) assumed to be homogeneous, monolingual, White spaces.

How do scores compare with region as a predictor in a simple regression model?

The final task I did was to run a simple regression model to see if instructional supports, language-specific supports, or overall resource supports differed as a function of region. The results of these models are reported in the following tables, with significant results marked by asterisks.

Notes about the table

Per common conventions:
* A single asterisk denotes a p-value meeting the .05 threshold.
* Double asterisks denote a p-value meeting the .01 threshold.
* A triple asterisk denotes a p-value that meets the .001 threshold.
* n.s. denotes a p-value that was not significant.

The first table includes the results of the instructional supports model, using the centered composite instructional score as the outcome variable with region as the only predictor. Instructional language services included:
* bilingual programs offered that cover core content areas
* two-way immersion or dual language programs offered that cover core content areas
* integrated English as a second-language (ESL) model offered
* pull-in/push-out ESL models offered
* English speaking educational paraprofessional(s) available
* bilingual educational paraprofessional(s) that speak the primary non-English language spoken by the majority of English learner-identified students available
* sheltered content classrooms used
* other instructional supports used

Table 1: Results of Instructional Supports Model

Table 2: Results of Language Supports Model

The second table includes the results of the additional language supports model, using the centered composite support score as the outcome variable with region as the only predictor.

Additional language-focused supports and services included:
* tutoring services
* summer programs
* credit recovery options
* mentoring
* distance education options
* other supports and services offered

Table 3: Results of Total Resources Model

The final table includes the results of the overall resources model which used the centered composite score for total resources offered as the outcome variable and region as the sole predictor.